Localization Techniques in Wireless NetworksPresented by: Rich Martin
Joint work with: David Madigan, Wade Trappe, Y. Chen, E. Elnahrawy, J. Francisco, X. Li,, K. Kleisouris, Y. Lim, B.
Turgut, many others.
Rutgers University
Presented at WINLAB, May 2006
2
• Technology trends creating cheap wireless communication in every computing device
• Radio offers localization opportunity in 2D and 3D • New capability compared to traditional communication networks
Motivation
3
A Solved Problem?
• Don’t we already know how to do this?– Many localization systems already exist
• Yes, they can localize, but ….– Missing the big picture – Not general
4
Open problem
• Analogy: Electronic communication 1960’s Leased lines ( problem solved! ) ->
1970’s Packet switching ->
1980’s internetworking ->
1990’s “The Internet”:
General purpose communication
• General purpose localization still open
5
Research Challenge
• General purpose localization analogous to general purpose communication.
• Work on any wireless device with little/no modification • Supports vast range of performance• Device always “knows where it is”• “Lost” --- no longer a concern
• Use only the existing communication infrastructure?– How much can we leverage? – If not, how general is it?– What are the cost/performance trade-offs?
6
Outline
• Motivation• Research Challenges• Background• General-purpose localization system• Open issues• Conclusions
7
Background: Localization Strategies
• Active– Measure a reflected signal
• Aggregate– Use constraints on many-course grained measurements.
• Scene matching– The best match on a previously constructed radio map– A classifier problem: “best” spot that matches the data
• Lateration and Angulation– Use distances, angles to landmarks to compute positions
8
Aggregate Approaches
• Formulations:– Nonlinear Optimization problem– Multi-Dimensional Scaling– Energy minimization, e.g. springs– Classifiers
• A field of nodes + Landmarks
• Local neighbor range or connectivity
[X2,Y2][X1,Y1]
[X3,Y3]
9
Scene Matching
• Build a radio map[X,Y,RSS1,RSS2,RSS3]
Training data
• Classifiers: Bayes’ rule
Max. Likelihood
Machine learning (SVM)
• Slow, error prone• Have to change when
environment changes
QuickTime™ and aTIFF (LZW) decompressor
are needed to see this picture.
dBm
Landmark 2
10
Lateration and Angulation
DD44
DD11
DD22
DD33
11
Observing Distances and Angles
• Received Signal Strength (RSS) to Distance– Path loss models
• RSS to Angle of Arrival (AoA)– Directional antenna models
• Time-of-Flight to distance(ToF)– Speed of light
12
RSS to Distance
RSS Fit
RSS to Distance
RS
S (
dB
m)
Distance (ft)
0 50 100 150 200
-100
-80
-60
-40
13
Time-of-Arrival to Distance
14
RS
S (
dB
m)
Angle (degrees)0
-80
-40
RSS to Angle
Actual
Smoothed
Modeled
15
Results Overview
• Last 6 years --- many,many varied efforts– Most are simulation, or trace-driven simulation
• Aggregate• 1/2 1-hop radio range typical.• Requires very dense networks (degree 6-8)
• Scene matching• 802.11, 802.15.4: Room/2-3m accuracy [Elnahrawy 04]• Need lots of training data
• Lateration and Angulation • 802.11, 802.15.4: Room/3-4m accuracy• Real deployments worse than theoretical models predict (1m)
16
Outline
• Motivation• Research Challenges• Background• General-purpose localization system• Open issues• Conclusions
17
General Purpose Localization
• Goal: Infrastructure for general-purpose localization
• Long running, on-line system – Weeks, months
• Experimentation• Data collection
18
Packet-level, Centralized Approach
• Deploy Landmarks– Monitor packet traffic at known positions– Observe packet radio properties
• Received Signal Strength (RSS)• Angle of Arrival (AoA)• Time of Arrival (ToA)• Phase Differential (PD)
• Server collects per-packet/bit properties– Saves packet information over time
• Solvers compute positions at time T– Can use multiple algorithms
• Clients contact server for positioning information
19
Software Components
ServerServer
Landmark1Landmark1
Solver1Solver1
ClientClient
Landmark2Landmark2
Landmark3Landmark3
Solver2Solver2
[PH,X1,Y1,RSS1]
[PH,X2,Y2,RSS2]
[PH,X3,Y3,RSS3]
PH
PH
PH
Headset?
[PH][X1,Y1,RSS1][X2,Y2,RSS2][X3,Y3,RSS3]
[XH,YH]
[XH,YH]
20
Award for Demo at TinyOS Technology Exchange III
21
Landmarks
• 802.11:– RSS – AoA– ToA
• 802.15.4 – RSS
• Future work: – Combo 802.11, 802.15.4– Reprogram radio boards, more accurate ToA – MIMO AoA?
22
Angle-of-Arrival Landmark
Rotating Directional Antenna
Reduces number of landmarks and training set needed to obtain good results
Does not improve absolute positioning accuracy (3m)
[Elnahrawy 06]
23
Localization Server
• Server maintains all info for the coordinate space– Spanning coordinate systems future work
• Protocols to landmarks, solver and clients are simple strings-over-sockets
• Multi-threaded Java implementation– State saved as flat files
24
Localization Solvers
• Winbugs solver [Madigan 04]
• Fast Bayesian Network solver [Kleisouris 06]
• Scene Matching Solver future work– Simple Point Matching– Area-Based Probability
25
Example Solver: Bayesian Graphical Models
X Y
D
S
Vertices = random variablesEdges = relationships
Example: Log-based signal strength propagation
Can encode arbitrary prior knowledge
€
D = (x − xb )2 + (y − yb )2
€
S = b1 + b2 log(D)
b2b1
26
Incorporating Angle-of-Arrival
S1 S2 S3 S4
D1 D2 D3 D4
YiXi
A1 A2 A3 A4
b11b01 b12 b13 b14b02 b03 b04
Position
Distance
Angle
RSS
Propagation Constants
Minus: no closed form solution for values of nodes
27
Computing the Probability Density using Sampling
28
Clients
• Text-only client• GUI client is future work
– CGI-scripts to contact server, update map– GRASS client– Google
29
Outline
• Motivation• Research Challenges• Background• General-purpose localization system• Open issues• Conclusions & Future Work
30
Open Issues
• Social Issues– Privacy, security
• Resources for communication vs. localization• Scalability
31
Social Issues
• Privacy – Who owns the position information?
• Person who owns the object, or the infrastructure?
– What are the “social contracts” between the parties?• Economic incentives?
– Centralized solutions make enforcing contracts and policies more tractable.
• Security– Attenuation/amplification attacks [Chen 2006]
– Tin foil, pringles can
– No/spoofed source headers?– Attack detection
32
Communication vs. Localization
• Resource use for Localization vs. Comm.? • Ideal landmark positions not the same as for comm.
coverage [Chen 2006]
33
Scalability
• Can scale to 10’s of unknowns in a few seconds
• Can we do 1000s? Localize 10 Points
0102030405060708090
M1 M2 M3 A1
Bayesian Networks
Time (secs)
slice wdWinBugs
34
Future Work
• Rebuild and deploy system– Gain experience running over weeks, months
• Continue to improve landmarks – High frequency, bit-level timestamps
• Scalability– Parallelize sampling algorithms
• Security– Attack detection– Algorithmic agreement
• Social issues?
35
Conclusions
• Time to defocus from algorithmic work • Localization of all radios will happen
– Expect variety of deployed systems– Demonstration of cost/performance tradeoffs
• Technical form, social issues not understood
36
References
• Today’s talks:– Kosta: Rapid sampling of Bayesian Networks– Yingying: Landmark placement
• E. Elnahrawy ,X. Li ,R. P. Martin, The Limits of Localization Using Signal Strength: A Comparative Study In Proceedings of the IEEE Conference on Sensor and Ad Hoc Communication Networks, SECON 2004
• D. Madigan , E. Elnahrawy ,R. P. Martin ,W. H. Ju ,P. Krishnan ,A. S. Krishnakumar, Bayesian Indoor Positioning Systems , INFOCOM 2005, March 2004
• Y. Chen, W. Trappe, R. P. Martin, The Robustness of Localization Algorithms to Signal Strength Attacks: A Comparative Study, DCOSS 2006